Advanced analytics for voice of the customer - SPA Future Thinking

Customer Experience
Management
Advanced Analytics for Voice of the Customer (VoC) Programmes
William Ullstein, SPA Future Thinking
Dan Hillyard, The Analytics Hub

Customer Experience Management
2
Although most organisations conduct
continuous customer satisfaction surveys, their
ability to deliver real insight and implement
actions to improve customer experience from
basic survey data is limited.
The other big challenges in retaining, improving
and understanding how to best manage
customer relationships are to move beyond the
raw survey and customer transactional data
and to understand the voice of the customer in
its true context.
However, integrating all streams of structured
and unstructured data are simply not efficient or
cost-effective without advanced analytics.
These approaches can help companies build
customer loyalty and profitability by:
• Contextualising customer insights
• Identifying areas for customer experience
improvements through new products and/or
services
• Building sophisticated models to
understand, predict and proactively manage
customer relations and interactions
The application of these advanced analytical
techniques enables users to extract knowledge
from data and to create predictive models for
better decision making.
A range of advanced analytical procedures
can be utilised and integrated into Voice of the
Customer programmes to better understand
customers including:

Key Driver Analysis
Modelling Analysis
Insight Mining
Forecasting / Target Setting
Data Linkage
Text Mining
Segmentation
Action Planning
Social Media Integration
Companies are increasingly challenged to drive loyalty and satisfaction to
improve the customer experience. While organisations have programmes
to measure and manage customer experiences, there are major challenges
in analysing the multiple information sources and the huge amount of
structured and unstructured data available.
Introduction

3
The traditional ‘quadrant’ style approach to
drivers analysis doesn’t always fully explain the
multitude of data in large customer satisfaction
surveys. Therefore a range of data exploration
and data modelling techniques are applied to
offer a more engaging deliverable.
There are a number of options available to
perform this task; however, the approach to the
process will involve measuring the association
between a set of predictors and an overall
indicator of success or performance. Details of
just some of the appropriate techniques include:
Regression analysis: The most common
technique with many different variants falling
under this umbrella term. Commonly, regression
models are made up of a number of predictor
variables that are combined to predict a
dependent or target variable (most commonly,
overall satisfaction). Additionally, the regression
model allows the user to gain an understanding
of the importance of each of the key drivers.
This analysis is designed to concentrate
management activities on those issues that
make the most difference.
Driver analysis encompasses a number of statistical tools and has the aim
of identifying the key issues that are driving overall customer satisfaction,
customer loyalty or recommendation. This allows organisations to focus
management schemes and issues that will elicit real results in terms of
improved satisfaction and improved loyalty.
Key Driver Analysis

4
Regression models can be based on a number
of predictor variables. Alternatively, the predictor
variables can be formed from factors/groups
derived from factor analysis or cluster analysis.
As regression analysis has some known
limitations, in that models can be affected
greatly by interactions between predictors,
more progressive techniques for identifying the
relative importance of predictors can be utilized.
Relative Importance identifies the comparative
worth of each predictor without the bias
associated with some traditional regression
techniques. The result is a more truthful set of
measures that are proportional and scale for
comparisons with equivalent analyses.
Further techniques that can be employed in key
driver analysis are discriminant analysis and
CHAID / decision trees.
Discriminant analysis is an extension of
regression techniques described above. Its
primary objective is to form a linear combination
of predictor variables that discriminate between
previously defined groups e.g. satisfied
customer vs. non-satisfied customers. This will
allow identification of key drivers that define the
inclusion on each of these groups.
CHAID and Decision Tree analysis are terms for
a number of algorithms that allow the creation
of classification systems, displayed in the
form of a decision tree. Component questions
within the questionnaire are used as predictor
variables.
This analysis seeks significantly homogenous
groups in the data set that will explain overall
satisfaction. Alternatively the dimensions from
a factor analysis can be used as predictor
variables.
Key Driver Analysis
Relative Importance identifies
the comparative worth of each
predictor without the bias
associated with some traditional
regression techniques.
The results from all of these
techniques can then be compiled
and assimilated to form a
definitive picture of what is driving
satisfaction and loyalty in the
customer populations.

5
A combination of driver analyses and modelling
techniques to explicate larger data sets that are
associated with continuous customer feedback
programmes can be utilized. This identifies
what is most critical for increased performance
and allows clients to ask what-if? and test
scenarios as an aid to planning of improvement
actions. These simulation models resulting
from regression analysis or similar techniques
will allow the user to run varied scenarios with
respect to changes in key drivers and view
resultant change in customer satisfaction
recommendation or other key metrics.
As this is intrinsically linked to key driver
analysis, modelling analysis can share
many of the techniques mentioned in the
previous section. However, the process
in building a model is to predict an overall
indicator of success or performance based
on a set of associated questions. In addition
to correlations, regression, discriminant and
decision trees; more radical and bespoke
methods may be used to model data. SEM and
Neural Nets are examples of more progressive
techniques used to model data.
The main deliverable in a modelling analysis
process is an interactive tool for the simulation
and testing of scenarios. Assessment of any
models developed will be communicated
along with the relevant assumptions or known
limitations pertaining to that analysis on
completion.
The sampling requirements for most of the
underlying methods in driver and modelling
analyses are such that these systems would
most likely be developed at a market level.
Dissemination to subgroups or an equivalent
can be produced if data is valid at this level.
Modelling Analysis
The main deliverable in a
modelling analysis process
is an interactive tool for the
simulation and testing of
scenarios.

6
A mix of methods and approaches drawn from
data mining, statistics and ‘standard’ research
analysis are employed. This will typically
generate a story, unearthing patterns and
differences within the data that other techniques
may overlook. This approach is increasingly
beneficial to larger tracking studies, especially
where top-line data is relatively ‘flat’.
Insight mining is a term used to describe the
application of data mining techniques to large
survey datasets to yield additional insight. The
development of the technique is driven from the
need help to uncover what the data is telling us.
The analysis involves a tailored investigation of
the basic data sets, which can begin with little
or no preconceived direction. A mix of methods
and approaches drawn from data mining,
statistics and research analysis are employed.
The techniques employed to conduct insight
mining include, but are not limited to; cross
tabulations, drill-down analysis, ANOVA,
hypothesis testing, correlations, CHAID,
regression techniques, text analytics, quadrant
plots, threshold analysis, waterfall charts and
geographical maps.
The visualisation of insight mining results is a
critical element to its successful communication.
Successful Insight Mining should generate a
story, unearthing patterns and differences within
the data.
Continuous customer experience surveys involve collecting a great deal of
structured data, and clients often need help to uncover what their data is
telling them. A method which can be employed is insight mining.
The analysis is a tailored investigation of the data set, which can begin with
little or no preconceived direction, or be directed towards an observed shift
in the data.
Insight Mining
Data mining for exploring the data and
identifying issues: Cross tabulations,
drill-down analysis, ANOVA,
hypothesis testing.
Association for testing and developing
relationships within the data:
Correlations, CHAID, regression
techniques, key driver analysis, text
analytics.
Visualisation for communicating
the findings: Quadrant plots,
threshold analysis, waterfall charts,
geographical maps.
Main examples of techniques
are summarised below:

7
By forecasting the current trend a base is
established from which to set future targets.
Additional supporting information, thresholds
and benchmarks can be integrated to help
understand the effect of current business issues
and initiatives to set and control future targets.
Forecasting commonly refers to the fitting
of a model or trend using obtained data and
extending the fit to future events or levels
outside of the current range of scoring. This
analysis can bridge a gap between modelling
and action planning, by predicting the natural
evolution of proceedings.
All methods used in forecasting are designed
to predict likely performance levels in unknown
circumstances. The process can involve
simplistic or complicated methods to carry out
predictions, and as such, there are appropriate
techniques for a given requirement available.
A range of sources can all form the analysis
from which more complex forecasts are
generated. These include:
• Volumetric sales forecasting
• Time series/econometric approaches for
trend data
• Social/market/technology/‘blue-sky’ trends
combining research
• Government or third-party statistics
Forecasting analysis can then use a range
of outputs to communicate predictions from
interactive graphics to portals for the prediction
of unknown or uncharted time periods and
scenarios.
The ability to forecast and set targets within continous tracker programmes
aids business planning and strategic direction.
At the modelling stage techniques to understand and simulate how data
behaves are applied. In forecasting models and simulations these are
extended to predict the natural progression of the data.
Forecasting / Target Setting
The process can involve
simplistic or complicated
methods to carry out
predictions, and as such, there
are appropriate techniques
available for a given
requirement

8
The overall aim of data linkage is to create links
and connections between separate surveys
or between surveys, customer databases and
other data. Analysis that incorporates measures
across several programmes and data streams
can yield a richer view of the overall customer
experience.
Database linking provides solutions involving
measures of association between survey and
non-survey data. It investigates how customer
feedback predicts actual customer transactional
behaviour or the creation of a balanced
scorecard or dashboard approach to monitoring
the state of your business.
Often the service issues affecting one stage of
a process, such as problems at new account
creation, can be tracked through to subsequent
stages in the customer journey. There are
strands that link performance levels across
many touch-points at which customers interact
with an organisation. There are constituents
that link performance levels across touch-
points at which customers interact with an
organisation.
The analysis incorporates measures across
several programmes and data and can
therefore yield a richer view of the overall
customer experience. The bulk of this work
is completed with database manipulation
association and cleaning tools and therefore
the deliverable is not as clearly defined as
previously mentioned analyses. Connected
sources of information are the main benefit
of this approach. This allows organisations to
conduct analysis of data with greater clarity and
more involvement outside of the sole customer
experience.
Data Linkage
Transactional customer feedback
surveys
Other survey data, e.g. exit interviews,
employee surveys, mystery shopping,
brand, lost sales, etc.
Customer data, e.g. transactions,
purchase behaviour, profiles, history, etc.
Non-survey data, e.g. revenue,
profitability, lifetime value, etc.
Operational measures, e.g. response
time to a complaint, etc.
Unsolicited feedback, e.g. social media,
blogs, comment cards, call center, etc.
Overview of potential data for
linkage analysis:

9
Text mining begins with natural language
processing (NLP) that enables computers to
discern meaning from human language. A
complete solution can automatically extract
categories, emerging themes, customer
sentiment and root cause of customer feedback
in a way that cannot be accessed using
traditional analytical techniques.
Whereas data mining extracts information from
structured databases, text mining extracts
information from unstructured data. There are
two approaches: linguistic and statistical.
The linguistic approach involves identifying
linguistic elements of language and structures
that relate them to each other. Those elements
are the keys to meaning. This is much like
parsing or diagramming a sentence to identify
parts of speech. If you have an adjective, for
example, what noun does it modify?
You don’t need human intervention to train
a classifier, but you do want to identify
parts of speech (which is done easily with
a conveniently packaged list known as a
“dictionary”); words that are commonly used by
people who are dissatisfied (which can come
from a dictionary or thesaurus); and information
about sentence structure (if you have two
nouns, which is the subject and which is the
object? Did John throw the ball or did the ball
throw John?).
Text mining is the discovery of information by analyzing text. Sources may
include any written form of communication captured electronically or verbal
communication transcribed into text. Insights gleaned from this text could
reflect the meaning of what the author intended or provide entirely new
information.
Text Mining
Whereas data mining extracts
information from structured
databases, text mining extracts
information from unstructured
data.

10
In the statistical approach, words and phrases
are treated as abstract objects. You use purely
their mathematical relationship to each other.
This approach most often involves machine-
learning. Is your customer dissatisfied? Is he
satisfied? Is another customer talking about
your latest product? Using a sample of the text
and assignments in a number of categories, the
computer scans them for common elements.
The machine learns by example based on
training data assigned by human beings. If
a business receives a million emails a day,
you would take a small sample—say, 500 to
2,000 emails—and manually classify them.
Then the computer would scan the sample to
identify relationships in the text that hold clues
for whether a particular email may be from a
contented customer. People using obscene
language tend to be unhappy, so simply
scanning for profanity in your sample can
distinguish email from irate customers.
There are many ways to train a classifier
without human intervention. One important
method is clustering, in which you look at what
words naturally go together, automatically
identifying common themes just by looking at
one or two emails, rather than 200,000, for
instance.
Text Mining
In the statistical approach,
words and phrases are treated
as abstract objects. You use
purely their mathematical
relationship to each other.

11
Segmentation enables companies to employ a more targeted approach,
where initiatives and priorities are tailored for distinct groups of customers.
Understanding how groups differ in terms of expectations and behaviour,
makes sure that the experience is not a one-size-fits-all approach.
Segmentation
Given its broad connotation, segmentation
is used to describe very simple to extremely
complex routines. Indeed segmenting data into
male and female, or the young and the old,
can often give a good implicit segmentation
in market research surveys. However, when
we use the term segmentation, we mean the
process of grouping respondents by their
situation, behaviours or attitudes in respect to a
service.
This process of classifying a market into distinct
subsets, so that customers within the subset
have similar needs or priorities, is useful in
understanding the data we have collected and
enables a business to tailor the service, so it is
applicable for all. The result is a means to focus
efforts in the right areas, to maximise the return
in service efforts. Understanding how groups
differ in terms of expectations and behaviour,
makes sure that the experience is not a one-
size-fits-all approach.

12
Segmentation
Before segmenting, extensive exploratory
analysis is often conducted. An existing
understanding of the behaviour and motivation
of customers is required, if we are to ask
the right questions. This can incorporate
quantitative and qualitative elements and
is often combined with a good deal of desk
research.
The process of creating segments most
commonly utilises multivariate techniques
to analyse responses to key elements from
a survey. These techniques essentially find
patterns in the data and use these patterns to
create groups of respondents, where there are
similarities between respondents in the same
cluster, and differences between respondents in
different clusters.
The procedure is iterative and multiple
solutions, using alternative techniques and sets
of questions, are produced. We also look at
how segmentations “evolve” as we increase the
number of segments. Interim stages, where the
survey data is used to create pseudo variables,
are often investigated as possible elements to
segment on. This can be useful if the survey
questions are less suitable for a successful
segmentation solution.
Most scenarios that are tested are eventually
discarded, in favour of a few most promising
solutions. Rigorous testing of segments is
undertaken to reduce the field of candidates to
the final few and hopefully the final one.
The chosen segmentation will typically be
that with optimum levels in terms of segment
volume and value, differentiation and
compatibility / accessibility. The requirement is
always an actionable strategy for an improved
performance and the final solution deployed will
be measured by this prerequisite.
Segmentation is perhaps one of the best
examples of an analytical procedure where
techniques are discussed as the critical aspect,
and this is perhaps overstated. In truth the
value in how the exploration is conducted
far outweighs the selection of an appropriate
technique. This is not to say that all techniques
give the same results, this is definitely not
the case, but the main aim of understanding
customer behaviour and motivations should be
kept in mind throughout.

13
This analysis will often be constructed at a
strategic or market level and then disseminated
to local level using an appropriate method
of transference. Alternatively it is possible to
construct a plan of action at the lowest level
and translate this to overall performance.
Action planning shares techniques with driver
analysis as it also attempts to quantify and
prioritise elements of the survey. The methods
used to support an action planning process
can include correlations, regression, relative
importance, hypothesis testing, gap analysis,
benchmarking and threshold analysis.
An element of action planning often discussed
is the setting of appropriate targets. The target
setting analysis is a multi-stage process which
takes into account background and supporting
information available outside of the customer
experience programme.
The current trend in scoring and likely forecast
can be used to quantify the target in terms of
an improvement from the natural progression
of the programme. Benchmarking within the
programme or against competitors combined
with analysis of performance thresholds helps
to justify targeted performance levels.
The overall aim of the target setting
methodology is to determine the correct shift
in a given parameter that will result in an
appropriate target for an associated reporting
period. The combination of key service areas
for prioritised action and targeted levels of
performance in key areas forms the main
deliverable in the action planning process.
Clarity in the formulation of such an action
planning process also becomes a critical article.
Action planning is from an analytical point of view a term used to describe
the practice of constructing a technique or process to identify main areas in
which to concentrate efforts.
Action Planning
The current trend in scoring
and likely forecast can be
used to quantify the target in
terms of an improvement from
the natural progression of the
programme.

14
Traditional quantitative customer satisfaction
tracking surveys are not set up to collect and
manage customer feedback across social
media. This channel is increasingly where
customer conversations are taking place. For
example Facebook has over one billion users,
Twitter has 600 million accounts. And the
number of reviews on review sites such as Yelp
has increased dramatically over recent years. It
is an unmediated conversation and companies
are scrambling to participate.
For organisations to continue to know what
customers think and how they feel about
products and services, companies must be
managing social feedback channels.
At a minimum, companies can listen to the
voice of the customers in less structured
forums. Even better, social feedback will be
integrated with surveys and other structured
and unstructured data for a complete picture of
the customer.
Social Media Integration
If an organisation wants to
continue to know what the
customers think and how
they feel about products and
services, companies must be
managing social feedback
channels.

15
This document discusses some of the
procedures in advanced analytics that can
be incorporated into customer experience
management programmes. This is, however,
part of a portfolio that is habitually changing
in its content and design. Although these
procedures change, understanding customer
feedback and the ability to grow information
from data will always be essential.
The segmentation section mentioned
procedures where techniques are often
discussed as the most critical aspect of
concern, and that this is erroneous. This is in
fact a danger in all the procedures we have
discussed and can often lead to narrowly
scoped investigations and poor findings.
Preparation and discussion are combatants
to poor analytical findings and the best results
are achieved when the vast majority of work is
exploratory.
Analytics form a powerful tool for understanding
the customer. Analytical systems can extract
knowledge from data that would otherwise
remain hidden, and the ability to create
predictive models for more informed decision
making is only possible because of the
advancement of statistical and data processing
tools.
That said, advanced analytics are no match for
a poorly conceived study or their use on data
that is not appropriate for the procedures and
techniques applied. It should always be kept in
mind that the most sophisticated multivariate
technique will only be as good as the data on
which it is based.
Conclusions
Understanding customer
feedback and the ability to
grow information from data will
always be essential.

16
SPA Future Thinking has developed to become one of the fastest growing and largest independent
market research companies in Europe, with more than 200 employees across offices in the UK,
France and Italy and a partner network in 38 countries worldwide. With combined industry experience
of over 40 years we offer thought leadership across a wide range of specialisms and we provide
genuine expertise and understanding across a wide range of industry sectors.
We provide strategic research and actionable insights to a wide range of leading companies; offering
both bespoke solutions, analytics and proprietary techniques to help guide the decisions that you
make in areas like, market development, new product development, innovation, customer satisfaction
and experience, brand, communications, buying behaviour and more.
www.spafuturethinking.com
The Analytics Hub was launched by SPA Future Thinking in September 2011 and offers data and
statistical expertise. The Analytics Hub offers a full range of analytic methodologies, including
segmentation, statistical modelling, dashboard and visualisation, text analytics, conjoint and discrete
choice analysis as well as forecasting and data mining, as part of its service. Our team has solid
credentials, with 40 years’ combined experience working on a wide range of research, marketing data
and forecasting projects.
www.theanalyticshub.com
About the Company

17
William Ullstein - Group Business Development Director at
SPA Future Thinking
A former Research Director, William is now responsible for new business
and marketing across the SPA Future Thinking group.
While a research practitioner, Will straddled the Quant / Qual divide
delivering commercially focused insight to a range of blue chip clients
including the likes of: British Gas, Carbon Trust, Serco and hibu (formerly
Yell).
Will was also the co-founder of The Sunday Times Best Green Companies
– research and analytics led awards, that sought to assess and celebrate
the CSR credentials of the UK’s greenest companies.
william.ullstein@spafuturethinking.com
+44 (0) 20 7843 9777
Dan Hillyard - Managing Director of The Analytics Hub
Dan is passionate about creating market segmentations that really
resonate with the marketers that use them and is an expert in forecasting
and choice-based conjoint techniques.
A natural communicator, Dan is happiest discussing the interpretation and
implications of analytics work and making the connections between data
and business issues.
Dan is also a graduate of University College London, a proud father and
maintains in the face of overwhelming evidence to the contrary that one
day he’ll make it as an obscure electronic music producer.
dan.hillyard@theanalyticshub.com
+44 (0) 1865 336 427
About the Authors

Advanced analytics for voice of the customer - SPA Future Thinking

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Empfohlen

Empfohlen (20)

Advanced analytics for voice of the customer - SPA Future Thinking